Comparison of neural networks and support vector machines applied to optimized features extracted from patients' speech signal for classification of vocal fold inflammation

The aim of this paper is to make an overall comparison between neural networks and support vector machines as two different types of artificial intelligence techniques applied to classify the most three widespread pathological factors as the consequences of vocal fold inflammation, known as vocal fold edema, nodules and polyp. As the analogous effects of these three pathological factors in changing the regular pattern of glottal opening and closure, influencing the speech signal quality, leads to a group of highly correlated extracted feature vectors, NN and SVM as nonlinear classifiers are used to achieve the best result in classification percentage. Besides, these powerful classifiers are supported with a genetic-based optimization algorithm to find the best and least correlated features which comprises their maximum final performance. Experiments on the basis of two different methods of feature extraction, wavelet packet sub-bands and Mel frequency scaled filter-banks, are carried out with some voiced signal. Test results taken from different NN's training methods and various SVM's kernels indicate the higher ability of the SVM for classifying these three types of vocal fold diseases, using entropy features which represent the vocal fold irregularities, with classification percentage of 94.12% vs. 73.53%